chore: use TypeScript parser instead of line matching in update_pw.mjs

[mxschmitt]

Summary

update_pw.mjs built the Monaco autocompletion types (frontend/src/components/Editor/types.txt) by editing the upstream .d.ts files line by line. This PR swaps that for edits based on an AST from the TypeScript 7 experimental API (typescript/unstable/*):

• stripFileHeader() → stripModuleSyntax(): parses each file and removes top-level import/import =/export … from statements whose module can't resolve in the concatenated file (relative paths and Node.js builtins). It also removes the file header (copyright, /// directives) but keeps any comment block directly attached to the first remaining statement.
• expect-types path rewrite: replaces the split('\n').map(line.replace(...)) with rewriteModuleSpecifiers(). This only rewrites string literals that are module specifiers (in import/export declarations, import = and import('…') types) and keeps the original quote style.
• All five files are parsed in one API session using an in-memory virtual FS (createVirtualFileSystem).
• Because the API is experimental, the script checks that every edited node's text matches what it expects at that offset. If the API ever reports offsets in a different unit, the script fails instead of writing a broken types.txt.
• Adds a small root package.json with typescript@^7.0.2 (the same version as frontend/), and updates CONTRIBUTING/AGENTS to run npm ci first.

Output changes

I diffed the old and new scripts on the same upstream files. The new script only adds back two comments that the line scanner was dropping by mistake:

• the JSDoc on Serializable (structs.d.ts); the old scanner treated it as part of the copyright header
• the URLPattern explanation comment (types.d.ts)

The @ts-ignored zod imports are kept, as before. Running tsc 7 over the old and new types.txt gives the same diagnostics.

The second commit regenerates types.txt (Playwright 1.64.0) with the new script. On top of #336 the only change is the two restored comments (5 lines).

Testing

• Ran updateFrontendTypes() with the old and new scripts and diffed the results (only the two comments above differ).
• Synthetic checks: /// + copyright header removal, // @ts-ignore kept on a remaining first import, specifier rewrite in import type, export * from and import('…') (other string literals left alone).
• tsc --noEmit on old and new output: same error set.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-08T21:33:47.742700Z	0fe07ba	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[flakiness-io]

Test Results Summary

This comment shows the latest test results for this pull request.

Commit	Status	Test Results	CI Runs
d9d87e9	✅ Passed	50 passed · 1 flaky	1 runs

⚠️ 1 flaky

⚠️ Ubuntu 24.04 · firefox e2e/tests/visual.spec.ts:137 › should handle platform core related features › should handle the timeout correctly

View full report

ℹ️ About Flakiness.io in GitHub

Your team has set up Flakiness.io pull request comments. This comment is updated when

• The first test results arrive for this pull request
• A later run reports more results for the same commit
• A new commit is pushed and tested

Flakiness.io rewrites the comment in place, so it always reflects the most recently tested commit.